Cursor is widening its ambitions beyond an AI code editor. The company has outlined a from-scratch AI model, a new Git platform called Origin, and a mobile app designed to keep software agents moving even when users are away from their main machine.
The announcements point to a broader product direction: Cursor wants to support not only code generation, but also the surrounding workflow where agents read repositories, write changes, respond to feedback, and recover from blocked tasks.
A self-trained model is now on the clock
Cursor shared new timing for its first fully self-trained AI model. Co-founder Michael said at the company’s event that training is underway, and the model should ship within the next few weeks.
The model is being trained from scratch. That is a change from earlier Composer models, which used an open-source base. The source article states that the new model is on par with Opus and GPT in size, and that it uses ten to twenty times more compute than previous Cursor models.
The collaboration with SpaceX on training the model was already known. The company behind Cursor, Anysphere, was recently acquired by SpaceX. What changed in this announcement is the clearer timeline and the emphasis that Cursor is building a model from the ground up.
Cursor also says the model is meant to work beyond coding. That matters because the company’s other announcements are not limited to writing code inside an editor. Origin and Cursor Mobile both focus on the operational side of AI-assisted work: coordinating agents, managing repositories, resolving blockers, and reviewing outputs.
Origin brings Git into an agent-heavy workflow
The second major announcement is Origin, a Git platform built for both humans and AI agents. Co-founder Tomas, who joined Cursor through the Graphite acquisition, said Origin runs on a new Git architecture built on top of cloud providers.
The goal is not just to host code. Origin is designed around the possibility that many agents may need to interact with the same repository at the same time. In load tests, the team simulated thousands of agents reading from and writing to a single repository simultaneously.
Cursor says Origin can handle several parts of the development loop that often slow down automated coding work:
- Resolving merge conflicts
- Fixing failed CI tests
- Handling comments
Those features are significant because AI agents can produce useful work but still get stuck in the practical mechanics of software delivery. A platform that can coordinate repository access, merge outcomes, test failures, and review comments is aimed at making agent work less isolated and more continuous.
Origin is already running internally and with select partners. Broad availability is planned for fall. That means Cursor is presenting the platform as more than a concept, while still keeping wider access for a later rollout.
Cursor Mobile puts agent control on iOS
The third announcement is Cursor Mobile, launching as an iOS beta. The app is built around remote management of agents rather than full desktop coding.
Cursor says the mobile app lets users manage agents remotely, unblock stuck tasks, and review or comment on screenshots generated by agents. A remote control feature also gives access to agents running locally.
That framing is important. Cursor Mobile is not described as a replacement for the main development environment. Instead, it extends control over agent workflows to situations where a user needs to check progress, give feedback, or clear a blocker from a phone.
In a workflow with multiple agents, that kind of remote visibility can matter. If an agent is waiting for direction, a user does not necessarily need to be at a desk to respond. If an agent produces a screenshot for review, the user can inspect and comment through the app.
Why the three announcements fit together
The model, Origin, and Cursor Mobile are separate products, but they describe one connected strategy. Cursor is building more of the stack needed for AI agents to participate in software work from start to finish.
The self-trained model addresses the intelligence layer. Origin addresses the repository and collaboration layer. Cursor Mobile addresses the control layer, giving users a way to supervise agents when they are not directly inside the desktop environment.
Cursor’s announcements also show how the company is thinking about scale. The model uses ten to twenty times more compute than previous Cursor models. Origin has been tested with thousands of agents working against a single repository. Cursor Mobile adds a way to keep those agents from sitting idle when human input is needed.
The common theme is continuity. Cursor is not only trying to make agents more capable; it is also trying to reduce the points where agent work stalls because of merge conflicts, CI failures, comments, screenshots, or the simple fact that a user is away from the keyboard.
What happens next
The clearest near-term item is the model. According to co-founder Michael, it should ship within the next few weeks. Because it is fully self-trained and built from scratch, it will be an important test of Cursor’s ability to move beyond models based on an open-source foundation.
Origin has a longer public timeline, with broad availability planned for fall. Until then, it is running internally and with select partners. Cursor Mobile is entering users’ hands through an iOS beta, giving the company another way to test how people want to supervise agents remotely.
Taken together, the announcements show Cursor positioning itself around AI agents as active participants in development workflows. The company is not only announcing a bigger model. It is also building the Git infrastructure and mobile controls needed to make agent-driven work easier to manage.